TY - GEN
T1 - EGGS
T2 - 2026 IEEE International Conference on Industrial Technology, ICIT 2026
AU - Gong, Gu
AU - Wang, Qiang
AU - Navarro-Alarcon, David
AU - He, Zhen
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Motion blur significantly degrades 3D reconstruction quality in actual scenarios. In this paper, we present EGGS (Event-Guided Gaussian Splatting), a framework that leverages events data to achieve sharp 3D reconstruction from motionblurred RGB images. By augmenting 3D Gaussian primitives with spatial-geometric event features extracted through multiscale analysis (16 × 16,8 × 8,4 × 4), we capture edge information unaffected by motion blur. The 512-dimensional event embeddings guide Gaussian optimization through feature consistency losses, ensuring accurate geometry despite degraded RGB inputs. Experiments on NeRF synthetic dataset show EGGS outperforms Ev3DGS by 1.75 dB PSNR on average across all scenes, while eliminating edge artifacts and preserving color fidelity. Our approach establishes event cameras as an effective complementary modality for robust 3D reconstruction.
AB - Motion blur significantly degrades 3D reconstruction quality in actual scenarios. In this paper, we present EGGS (Event-Guided Gaussian Splatting), a framework that leverages events data to achieve sharp 3D reconstruction from motionblurred RGB images. By augmenting 3D Gaussian primitives with spatial-geometric event features extracted through multiscale analysis (16 × 16,8 × 8,4 × 4), we capture edge information unaffected by motion blur. The 512-dimensional event embeddings guide Gaussian optimization through feature consistency losses, ensuring accurate geometry despite degraded RGB inputs. Experiments on NeRF synthetic dataset show EGGS outperforms Ev3DGS by 1.75 dB PSNR on average across all scenes, while eliminating edge artifacts and preserving color fidelity. Our approach establishes event cameras as an effective complementary modality for robust 3D reconstruction.
KW - 3D Gaussian Splatting
KW - Event-Guided Gaussian Splatting
KW - Event-based vision
KW - Motion Deblurring
UR - https://www.scopus.com/pages/publications/105038427890
U2 - 10.1109/ICIT64854.2026.11491451
DO - 10.1109/ICIT64854.2026.11491451
M3 - 会议稿件
AN - SCOPUS:105038427890
T3 - Proceedings of the IEEE International Conference on Industrial Technology
BT - 2026 IEEE International Conference on Industrial Technology, ICIT 2026
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 4 March 2026 through 6 March 2026
ER -